It’s undeniable. GenAI continues to make inroads in healthcare and life sciences. Last week, OpenAI announced OpenAI for Healthcare, including ChatGPT for Healthcare, and launched ChatGPT Health as a separate health and wellness experience. Now, Anthropic announced that it was expanding Claude into healthcare and downstream life sciences. It is also adding HIPAA-ready infrastructure plus new connectors aimed at prior authorization, billing support, clinical trials and regulatory work.
Anthropic says it is wiring Claude into the administrative layer of healthcare, with connectors aimed at coverage checks, coding and provider verification, plus PubMed for literature search. In life sciences, the new integrations shift downstream into trial operations and regulatory work, pulling in systems like Medidata and ClinicalTrials.gov along with preprints and drug-target databases. Separately, Anthropic is rolling out optional consumer-facing integrations for US Pro and Max users, including Function Health and HealthEx in beta, with Apple Health and Android Health Connect rolling out in beta this week. New agent skills include sample workflows for prior authorization and clinical trial protocol drafting.
Customers include Novo Nordisk, which Anthropic notes cut clinical documentation timelines from 12 weeks to 10 minutes. Banner Health reports 85% of Claude users working faster with improved accuracy, with more than 22,000 clinical providers on the platform. Other customers include Eli Lilly, AbbVie, Genmab, Sanofi, Elevance and Blue Cross Blue Shield.
The launch comes as AI in healthcare faces broader scrutiny. Anthropic’s Claude model powers FDA’s Elsa tool, which CNN reported has hallucinated nonexistent studies. Meanwhile, class action litigation against UnitedHealthcare over AI-driven claims denials is proceeding in federal court. A 2025 Penn LDI study found LLMs often disagree with themselves on clinical recommendations, though that research used earlier model versions.
Anthropic’s internal benchmarks show Claude Opus 4.5 achieving 92.3% accuracy on medical calculations and 61.3% on complex agentic medical tasks, a substantial improvement over earlier models. But operational context matters: 92.3% means roughly one wrong answer in 13 calculations. In medication dosing or risk scoring, the acceptable error rate with validated calculators is effectively zero. The 61.3% figure on complex tasks, a 38.7% failure rate on multi-step scenarios, suggests a capable drafting assistant that requires review, not an autonomous agent.
Independent physician-level comparisons don’t exist for these benchmarks. In clinical practice, “physician-level” calculation accuracy typically means calculator-level, since clinicians use deterministic EHR tools precisely because humans are fallible under time pressure.
Anthropic’s own demo materials include extensive disclaimers requiring biostatistician validation, clinical expert review, IRB approval and legal sign-off before any AI-generated protocol can be used. The disclaimers are a tacit acknowledgment that these tools are intended to accelerate drafting rather than decision-making.
The tools are generally available to Claude Pro, Max, Teams and Enterprise subscribers. Anthropic says Claude is also accessible via AWS Bedrock, Google Cloud Vertex AI and Microsoft Foundry, with systems integrators including Deloitte, Accenture and PwC supporting enterprise deployments.
